An Intelligent Intrusion Detection Using Deep Learning on CICIDS2018 for Cloud Security

Abstract

By enabling on-demand application use, computation, and data storage across the internet, cloud computing has moved to the front of modern digital infrastructure. Its distributed and dynamic nature makes it susceptible to various security risks, including anomalies and malicious traffic. As cyberattacks develop increasingly sophisticated, conventional detection technologies are not always up to the task of delivering quick and reliable responses. Still, this work provides a solid framework for detecting aberrant and hazardous network traffic in cloud settings using deep learning. The suggested method relies on the CICIDS2018 dataset, which contains information on actual network traffic as well as several cyberattack scenarios. To handle class imbalance, the proposed method uses a full data preparation pipeline including data cleaning, normalisation, and SMOTE (Synthetic Minority Over-sampling Technique) application. There are also robust feature selection methods like Random Forest and Symmetric Uncertainty. The methods guarantee that the pertinent, high-quality input data utilised to train the model is of high quality. The data on network traffic is then automatically learnt by a Convolutional Neural Network (CNN) model. Techniques like routine normalisation and dropout help to enhance the model's performance and stop overfitting. The model demonstrates excellent performance with a training accuracy of 99.85% and a testing accuracy of 99.76%. This approach takes into account the necessity for scalability and agility in cloud infrastructures as they evolve, therefore providing excellent detection accuracy. Intelligent intrusion detection systems (IDS) tailored to contexts powered by Industry 4.0 and the Internet of Things (IoT) have advanced significantly with this development.